Run models

Model family

Does Kling AI allow NSFW? Eight versions, tested

By the SpicyAPI team · Updated · Test data as of · How we test

Short answer

Mostly no. None of the eight Kling versions we tested delivered even two thirds of its runs; the newest, Kling 4.0, delivered 3 of 8, and Kling O3 came closest at 8 of 15. Kling’s problem is not quality: several versions post blind-test scores near the top of the video board. It is filtering: not one of the eight versions delivered a single run at the Moderate level, the second of five.

  • 8 versions tested
  • 0 delivered every run at all five levels
  • 72 test runs counted

Every version, side by side

VersionMild → Maximumdelivered / runsDeliveredSoftenedRefusedBlind testFromLink
Kling 4.0
  1. 2/2
  2. 0/1
  3. 0/1
  4. 0/1
  5. 1/3
3/81470.9$0.11/sec at 720pRun
Kling 3.0
  1. 7/7
  2. 0/2
  3. 0/3
  4. 0/2
  5. 0/1
7/150863.5$0.067/sec at 720pRun
Kling O3
  1. 8/8
  2. 0/4
  3. 0/3
  4. —
  5. —
8/150761.9$0.067/sec at 720pRun
Kling 3.0 Turbo
  1. 1/2
  2. 0/2
  3. —
  4. —
  5. 1/1
2/50356.8$0.11/sec at 720pRun
Kling 2.6
  1. 5/7
  2. 0/2
  3. 0/2
  4. —
  5. 0/2
5/130864.7$0.069/sec at 720pRun
Kling 2.5 Turbo
  1. 1/2
  2. 0/2
  3. —
  4. 0/1
  5. 1/1
2/60464.8$0.056/secRun
Kling 2.1
  1. 1/2
  2. 0/1
  3. —
  4. —
  5. 0/1
1/40371.8$0.035/sec at 720pRun
Kling 2.1 Master
  1. 2/3
  2. 0/2
  3. —
  4. —
  5. 0/1
2/60455$0.22/secRun

A dash means that level has not been tested yet. Softened runs returned something milder than requested; refused runs returned nothing. Data as of 2026-11-03; see how we test.

Good video, firm filter

Kling 2.1 and Kling 4.0 score 71.8 and 70.9 in the blind test, level with Wan 3.0. What pulls their Spicy Index down is the Freedom Score: refusals rather than softened runs make up most of the misses, so a request either goes through or stops entirely.

Sample sizes are small for several versions, and some have levels we have not tested yet. Read the table as a consistent pattern rather than precise rates.

What to use instead

If you came to Kling for its motion quality, Wan 3.0 matches its blind-test score and delivered all 51 runs. Seedance 2.5 delivered 45 of 46 with the highest Spicy Index outside the Wan family, and MiniMax H3 with LoRA delivered 58 of 60 while letting you load a character LoRA.

Where Kling still fits

For work Kling’s official app allows, it remains a strong choice, and Kling 4.0 and O3 add video editing that few other models offer. Keep it for that, and route content it refuses to one of the models above.

Questions

Is Kling 4.0 uncensored?

No. In our tests it delivered 3 of 8 runs, softened 1 and refused 4.

Which Kling version is least filtered?

Kling O3 delivered 8 of 15, the highest share among Kling versions, but it has levels we have not tested yet.

What is a good Kling alternative for NSFW?

Wan 3.0 delivered all 51 runs with a blind-test score level with Kling 4.0.